T HE RELATIONSHIP BETWEEN MEMORY SPECIFICITY AND CUE CONTENT

As mentioned above, we hypothesised that (1) the more AMT cues tap into a domain that is highly relevant for the respondent, the lower the proportion of specific memories retrieved by the respondent will be, and (2) the more AMT cues are experienced as discrepant with one’s actual state, the lower the proportion of specific memories retrieved will be. To test these hypotheses, we investigated the correlations between the proportion of specific and categoric memories retrieved in response to the AMT cues and the SDQ indices computed as described in the Analyses section. At first sight, the indices are not associated with memory specificity; IrT nor IdT correlated significantly with the proportion of specific memories, r = -.11, r = .17, ps = ns, respectively. There were no associations with the proportion of categoric memories either, r = -.02, r = -.06, ps = ns, respectively.

Table 6 Correlations between the proportion of specific and categoric memories and

between the indices expressing the self-relevance (IrT) and self-discrepancy (IdT) of the AMT % S % GC IrT IdT % S - -.63** .03 .40 % GC -.76** - -.05 -.19 IrT -.58 .07 - .29 IdT -.89** .57 .78** -

Data above the diagonal relate to the non-depressed patients (n = 22) or our sample; Data below the diagonal relate to our depressed subsample (n = 11).

%S = proportion specific memories retrieved during AMT administration; %GC = proportion general categoric memories retrieved during AMT administration; IrT = index expressing relevance of AMT in relation to all participants’ self-guides produced during SDQ assessment; IdT = index expressing relevance of AMT in relation to all participants’ self- guides produced during SDQ administration, also taken into account the discrepancies indicated by the participant.

- 69 -

Crane et al. (2007) found that cue content only related to memory specificity in remitted depressed patients, and Kremers et al. (2004; 2006a) found that a depressed subgroup of their borderline patients retrieved less specific memories compared to the non-depressed borderlines they examined. Inspired by these findings, we recalculated these correlations in the depressed and the non-depressed subgroups of our sample. These correlations are shown in Table 6. The correlations of the non-depressed subgroup are shown above the diagonal; under the diagonal the correlations of the depressed subgroup are displayed. In spite of the small selection – only 11 of our participants were currently depressed –, we now found that IdT was negatively and highly associated with memory specificity in the depressed subsample, r = -.89, p < .01, suggesting that depressed BPD patients have more difficulties retrieving specific memories as cues are more referring to self-discrepant domains. In marked contrast, this relation was reversed (but not significant) in the non-depressed participants, r = .40, p = .07. These correlations differed significantly, z = -4.38, p < .00111.

Although not significant, we also found a positive association in the depressed subsample between IdT and the proportion of categoric memories, r = .57, p = .065, (significantly different from the corresponding r = -.19, p = .39 in the non-depressed participants, z = -1.99, p < .05). Furthermore, a negative association was found between

11

This result was corroborated by a multiple hierarchical regression analysis predicting the (standardised) proportion of specific memories. Independent variables were all standardised. We entered depression severity on the first step, IdT and depressive status (non-depressed vs depressed) on the second step, and the interaction of IdT and depressive status on the third step. The overall model was significant, R² = .46, F(4, 27) = 5.67, p < .005, with no significant effect of IdT and depressive status, t(27) = .34, p = .74, β = .06, and t(27) = .84, p = .41, β = .13 respectively. The interaction of IdT x depressive status, on the other hand, was held back as a significant predictor, t(27) = -2.46, p = .02, β = -.42, as was depression severity, t(27) = -2.20, p = .04, β = -.37, pointing out that the interaction between depressive status and IdT is the most important determinant in this regression analysis, even when controlled for depression severity scores.

- 70 -

IrT and the proportion of specific memories retrieved, r = -.58, p = .064. This final correlation does not differ from the corresponding r = -.03 in the non-depressed subsample, z = -1.62, p = .10. No association was found in either subsample between IrT and the proportion of categoric memories retrieved, r = -.05 for the non-depressed participants, and r = -.07 for the depressed participants respectively, both p = ns.

IrT and IdT were found to be unrelated to depression severity in the overall sample, r = .05, r = .14, ps > .44 respectively, as well as in the non-depressed participants, r = -.17, r = -.02, ps > .45 respectively. In the currently depressed participants, IdT related to depression severity, r = .71, p < .02, but IrT did not, r = .53, p = .10.

DISCUSSION

The aims of this study were threefold. First, we were interested to see what relationships could be found between memory specificity and depression (severity) in BPD patients. Second, we examined the relationship between memory specificity and rumination in BPD. Finally, based on recent theorising and empirical findings on the role of cue content in the retrieval process, we investigated whether memory specificity is influenced by the personal relevancy of AMT cues.

Regarding the associations between memory specificity and depression severity, and between memory specificity and rumination, we found that both depression severity and rumination were negatively associated with memory specificity in BPD patients, suggesting that the more severe one’s depressive symptoms are, and the more one ruminates, the less one is capable of retrieving specific autobiographical memories. However, it should be noted that rumination and memory specificity were no longer significantly related when depression severity scores were partialled out. Furthermore, memory specificity and its relation with depression severity were independent of depressive state, and no direct association between memory specificity and depressive diagnosis was found. These findings suggest that depression severity is a more

- 71 -

important determinant of memory specificity in our BPD sample than rumination or depressive state.

Although our results do not deviate from what is generally found in depressed patients (e.g., Raes et al., 2005; Raes, Hermans, Williams, Beyers, et al. 2006; van Minnen et al., 2005), they are inconsistent with previous work that focused on the relation between memory specificity and depression (severity) in BPD (Arntz et al., 2002; Jones et al., 1999; Kremers et al., 2004; 2006a; Maurex et al., 2010; Reid & Startup, 2010; Renneberg et al., 2005). If any relation was found, memory specificity in BPD patients was related to depressive state (Arntz et al., 2002; Kremers et al., 2004; 2006a), but never to depression severity. Clear-cut explanations for our findings are lacking, though we would like to point out some considerations. First, the currently depressed and the non-depressed participants did not differ on any of the variables of interest (depression severity, memory specificity, rumination). Possibly, our research design may not have encountered important variables that either directly or indirectly (via depression severity) have influenced memory specificity. As mentioned before, differences in RMS between patients with and patients without BPD have found to be mediated by differences in IQ and education (Reid & Startup, 2010), and associations are reported with poor social-problem solving capacity (Kremers et al., 2006b; Maurex et al., 2010; Reid, 2008) and with less parasuicidal acts (Startup et al., 2001). Executive deficits or post-traumatic symptoms can therefore be considered as important missing variables. Second, but closely relating to the previous consideration, we should consider the possibility that depression in BPD differs in nature from depression in MDD. Possibly, depression in BPD should be considered in terms of affective instability following traumatic experiences. Finally, we should question why only a relatively small number of participants was acknowledged as currently depressed by the psychiatrist. Based on the cut-off scores mentioned in the BDI-II-NL Manual (van der Does, 2002) we can conclude that 78.1% of our respondents were considered (moderately) severely depressed, and that reported depression severity scores were high in the total sample

- 72 -

(M = 33.31; SD = 12.97). Measuring depression severity implicates measuring the absence or presence of symptoms, and vice versa. We could therefore wonder what exactly determines whether a psychiatrist judges someone as currently depressed or not. The use of a categorical diagnostic classification system as the DSM-IV (APA, 1994) may have some limitations on this behalf, since a diagnosis is made by judging the presence or absence of certain criteria, possibly without taking fully into account the severity of the total tableau of symptoms.

Regarding the relations between cue content and memory specificity, we expected participants to report less specific memories when AMT cues more closely relate to highly discrepant personal domains (high discrepancy) or to highly personal themes (high relevance). As for ‘high discrepancy’, we did not find a relation with memory specificity for the total sample. However, the currently depressed BPD patients in our sample did show a significant negative association between memory specificity and the extent to which the AMT cues referred to discrepant domains about the self (IdT). Furthermore, a nearly significant positive correlation between IdT and the proportion of categoric memories was found in this subsample. These correlations differed significantly from the corresponding correlations in the non-depressed participants. In addition, these findings were unrelated to depression severity. This pattern of results suggests that the more AMT cues are experienced as discrepant with one’s actual self- perceptions, the more difficult it is to retrieve specific memories on that AMT, at least when one is currently depressed. As for ‘high relevancy’, we only observed a marginally significant negative association with memory specificity for the currently depressed BPD-patients in our sample, and this correlation did not differ from the corresponding correlation in the non-depressed participants. No associations were found between cue relevance and the proportion of categoric memories retrieved.

It has been proposed that the activation of schemas that are discrepant with actual self- schemas would demand a reallocation of resources in an attempt to reduce the present discrepancies, thereby hindering the retrieval of specific autobiographical information

- 73 -

(Conway et al., 2004; Dalgleish et al., 2003). As mentioned before, BPD patients are prone to rumination (Smith et al., 2006), which could be considered as a discrepancy- based thinking, directed at reducing the discrepancies (Crane et al., 2007). In addition, rumination has often been associated with RMS (e.g., see Debeer et al., 2009; Raes, Hermans, Williams, Beyers et al., 2006). Furthermore, BPD patients are typically characterised by an instable sense of self, suggesting that their WS processes are mostly involved in self-coherence rather than in adaptive correspondence (Parker, Boldero, & Bell, 2006). We therefore would expect RMS in (all) BPD patients. However, we only found significant associations between discrepancy and memory specificity in our depressed participants. Nevertheless, the non-depressed participants probably have some ideals or standards they try to attain as well. And given the high co- morbidity rates with depression in BPD, it is unlikely that none of the non-depressed participants had never been depressed before, even though we do not have precise data on the number of previous episodes of depression of our participants.

The SMS (Conway, 2005; Conway & Pleydell-Pearce, 2000; Conway et al., 2004), and the CaR-FA-X model (Williams et al., 2007), however, also take into account executive functioning and traumatic experiences. Depression and rumination are known to be associated with lower levels of executive functioning, which could explain why we only found a relation between discrepancy and memory specificity in our depressed participants. Furthermore, as mentioned before, depression in BPD patients could be different in nature since it can be considered in terms of affective instability following traumatic experiences, or as a post-traumatic symptom. Unfortunately no measures on (post-)trauma (symptoms) nor on executive functioning were administered.

An alternative, though speculative, possible explanation is delivered by Parker et al. (2006), who suggested that BPD patients differed in self-complexity. They assume that high self-complex persons will abandon to focus on discrepant ideal-self content, thereby protecting them to develop depression. We could therefore (tentatively) hypothesise that our currently depressed participants would be low in self-complexity,

- 74 -

therefore focussing more on ideal-actual discrepancies, which would lead to less specific memories. However, since this hypothesis is speculative, further research is necessary to test it.

Three topics need further discussion. First, we need to analyse the different findings concerning relevancy, which we found was unrelated to RMS, and discrepancy, that was highly associated with RMS in depressed participants. According to the SMS (Conway, 2005; Conway & Pleydell-Pearce, 2000; Conway et al., 2004) an emphasis on discrepancies (= highly discrepant) regarding personal (= highly relevant) goals could lead to a reallocation of the resources in the disadvantage of autobiographical memory specificity, suggesting that relevancy as well as discrepancy are both necessary elements in influencing memory specificity, and none of them is sufficient as such. Our data do confirm that relevancy as such is not sufficient, since it does not lead to problems in retrieving specific memories. Since high discrepancy indices are related to difficulties in retrieving specific autobiographical memories, we can conclude that discrepancy as such is a necessary factor in the correlation. However, it is not clear whether discrepancy is a sufficient factor, since our discrepancy indices are constructed by multiplying previously determined relevancy scores (‘synonymy scores’) by the discrepancy scores, thereby taking into account relevancy as well. This was a necessary step, because we needed to determine to what degree AMT cues and self-reported self-guides were related. In order to clearly distinguish between relevancy and discrepancy, pure discrepancy scores should be obtained in future research. This can be done by constructing personal AMT’s, based on one’s self-reported self-guides, or, the other way around, by asking participants to what degree they think they ought to have, ideally would like to have, or fear to have the characteristics expressed by the AMT cues, and to what degree they actually believe they possess each of these characteristics at the moment. Tentatively, we do believe that relevancy as well as discrepancy are necessary factors, since we think that highly discrepant domains are

- 75 -

not necessary highly relevant for the respondent, thereby having less impact on one’s resources and thinking processes.

Second, it would be interesting to see whether memory specificity in (depressed) BPD patients depends on the sort of discrepancies they mainly report, given the fact that BPD patients often complain about disturbed mood and anxiety. Following the ideas of Higgins and co-workers (e.g., Higgins et al., 1985; 1986), we would expect that depressed BPD patients experience more discomfort following the confrontation with ideal self-guides, which would lead to less specific retrieved memories. Despite our small sample size, we conducted some preliminary analyses, using indices that only took into account the ought, the ideal, and the feared self-guides respectively. Stressing that a larger sample size is recommended for these analyses, our data suggest that, contrary to the expectations, discrepant ideas about one’s ought self are more important in relation to RMS than discrepant or relevant ideas about any other kind of self, and only in the depressed subgroup of our sample. Further research has to be conducted, of course, but such data could also help pointing out whether rumination content (as stressed in the definition of rumination by Nolen-Hoeksema, 1991) or its process characteristics (“negative, uncontrollable and perseverative”, as stated by Watkins, 2008), or both are influencing RMS.

Finally, it should be questioned why discrepancy scores were only related to depression severity in our depressed participants, but not in general. Again, no clear-cut explanations can be offered, although we could again assume that depression in BPD differs in nature from depression in MDD. It is conceivable that traumatic experiences in vulnerable subjects install a range of post-traumatic symptoms, including an increased attention to self-coherence and depressed state with associated (ruminative) information processes. The focus on self-coherence and the depressed state may be maintained by worry-like thoughts, considering how one should be or act, given the (disturbed) circumstances (ought discrepancies). All these processes use available executive resources which in turn are unavailable for guided search processes. Future

- 76 -

research should therefore also focus on post-trauma symptoms and executive functioning.

Our findings are not without limitations. As mentioned before, we did not collect information on the number of previous depressive episodes, (severity of) past traumatic experiences and post-traumatic stress. Furthermore, we failed to collect information about the medication use of our participants, as well as about their executive functioning. Future research should take these variables, of which is known that they could explain (additional) variance in memory specificity, into account in order to conduct more complete analyses.

Furthermore, we used a relatively small sample (n=34) and participants were included in our study based on a diagnosis made by the treating psychiatrist. No more valid and reliable procedures (e.g., SCID-I, SCID-II) were used, nor did we assess participants’ precise borderline features or self-complexity. Therefore, we cannot exclude that our findings are the artefact of a selection bias in the pool of BPD phenotypes. Thus, replication with proper selection instruments in a larger sample is recommended.

In addition, although the (written) AMT we used has been successfully used in previous research (e.g., Henderson et al., 2002; Raes et al., 2008; 2009), and analogous results were found in comparable samples, we have to admit that no direct comparison with the standard (oral) AMT has been done so far. The version we used differs procedurally from the standard AMT (no time constraints, taking into account the respondent’s own judgement regarding specificity or overgenerality, possible difficulties deriving the level of specificity from written answers that had no further comment), which may have influenced our measure of memory specificity, and therefore our results.

Further comments concern time span of data collection. AMT administration immediately followed SDQ administration. We cannot exclude that filling out the SDQ might have activated important schemas that could have affected memory retrieval during AMT administration. In future studies, self-guides should ideally be assessed at

- 77 -

least one week before running the AMT (Williams, personal communication, Berlin, October 1, 2009).

Finally, we need to be careful using the indices we created to test our hypotheses on the relations between self-relevance/self-discrepancy and memory specificity. Although a large sample of volunteers scored to what degree self-descriptions and AMT cues were synonymous, each volunteer only got a random selection of 67 matches, and each match was only scored by 4.11 volunteers on average. The raters were recruited using different ways (personal contacts of the first author, contacts of personal contacts, website advertisements on forums), which makes it harder to assess the homogeneity of this group. We did however include some descriptive variables, which show that the raters differ in many aspects from the participants. The raters had higher levels of education, were more often married and they usually were employed.

However, and notwithstanding the above limitations, the present results suggest that depression (in interaction with cue relevance) as well as depression severity play an important role in RMS in BPD patients, extending earlier observations by, for example, Kremers et al. (2004, 2006a), but contradicting other findings by Arntz et al. (2002), Jones et al. (1999), Kremers et al. (2004, 2006a), Maurex et al. (2010), Reid and Startup (2010), and Renneberg et al. (2005). Also, the findings concerning the importance of cue content are consistent with prior work in the field (e.g., Barnhofer et al., 2007; Crane et al., 2007; Spinhoven et al., 2007) and add to the preliminary evidence suggesting an important association between RMS and the extent to which cues activate highly (discrepant) personal domains, which fits with the influential SMS model on autobiographical memory by Conway and colleagues (Conway, 2005; Conway & Pleydell-Pearce, 2000; Conway et al., 2004). Nevertheless, we may have neglected to include potential important variables, such as history of trauma, data on the personal coping with or consequences of trauma, or measures of executive functioning that could account to a better understanding of RMS in BPD.

- 78 -

REFERENCES

American Psychiatric Association (1994). Diagnostic and Statistical Manual of Mental Disorders (4th ed.). Washington DC: Author.

Arntz, A., Meeren, M., & Wessel, I. (2002). No evidence for overgeneral memories in borderline